Fable 5.2 explained: when the top Claude model is worth it

Fable 5.2 is the executive chef of the lineup, built for the hardest and most ambitious technical work. It costs about forty dollars per million tokens written, double Opus 5.5. That sounds expensive, but price per token is not price per job.

Why the sticker price can mislead

Your bill is tokens times price per token. A stronger model may reach the answer in fewer attempts and fewer tokens. On a hard coding test, Fable 5.2 at high thinking matched Opus 5.5 at max (about 45%) for roughly half the cost per task.

Good fits

• Complex, multi-step engineering problems
• Tasks where cheaper models keep failing after more thinking
• Work where a wrong answer is costly

Poor fits

• Sorting, tagging and other high-volume simple work (Haiku 5.5)
• Clear, well-defined jobs (Sonnet 5.5)
• Everyday complex work that Opus 5.5 already handles

How to adopt it sensibly

1. Start with Opus 5.5 at medium thinking.
2. Let the model check its own work.
3. Raise thinking one level.
4. Only then move up to Fable 5.2, and start it at a lower thinking level.

Measure before you commit

Run one real task on both models and compare cost per finished job. Results vary by task, so treat published numbers as a guide, not a promise. For help with model selection as part of a broader AI strategy, talk to our team.